Gary Marcus

Rank 13 of 47
|
Score 83
5/18/2025, 12:05:34 AM
X
In reply to:
Colin Fraser
@colin_fraser
·
240d
@littmath @GaryMarcus @MelMitchell1 @rao2z That, I theorize, is why o3 can’t do this but it can do like algebraic geometry or whatever. This involves checking many cases, thinking steps ahead for each one, backtracking, and so on. It’s easy work but it’s work and there’s a lot of it.
Colin Fraser
@colin_fraser
·
240d
@rapid_rar @irl_danB @CHaesemeyer No. Maybe if it thought for more than 3 minutes it would have found a better move.
Colin Fraser
@colin_fraser
·
240d
@littmath @GaryMarcus @MelMitchell1 @rao2z I think we’ve talked about this before and you disagreed with me but I’ll ask again; I’ve found the hallmark of problems that they do badly at is they involve a lot of “busywork”—trying many cases, finding the correct path out of many possible options, etc. Do you find this?
Daniel Litt
@littmath
·
240d
@GaryMarcus @colin_fraser @MelMitchell1 @rao2z Let me put it this way—they can very reliably solve some out-of-distribution hard problems that humans can’t reliably solve. Being able to solve *all* problems is not a pre-condition for intelligence; otherwise people aren’t intelligent.
Gary Marcus
@GaryMarcus
·
240d
@littmath @colin_fraser @MelMitchell1 @rao2z Some, but not reliably. They can’t even solve large multiplication problems reliably.

LLMs exceed their predecessors but demonstrably still have not solved this problem, which is most clear when the training set is known, and when direct data augmentation is not possible.

Some
Daniel Litt
@littmath
·
240d
@GaryMarcus @colin_fraser @MelMitchell1 @rao2z They’re clearly generalizing out of distribution. E.g. they can solve some novel math problems.
Gary Marcus
@GaryMarcus
·
240d
@littmath New situations is clearly the weakness in these approaches, and has been forever.

@colin_fraser has documented this endlessly, informally; I gave the intuitions behind that in 1998. The recent Apple reasoning paper, the studies of @MelMitchell1 and @rao2z etc reinforce it.

The

The statement is part of a technical discussion about the capabilities of AI models, specifically their planning abilities and problem-solving skills. It critiques the models' performance in tasks requiring strategic planning, such as playing tic-tac-toe, and aligns with previous claims about their limitations.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement does not cause harm and is focused on technical critique. [+1]
  2. Principle 3:
    I will use my words and actions to promote understanding, empathy, and compassion.
    It promotes understanding by engaging in a detailed discussion about AI capabilities. [+1]
  3. Principle 4:
    I will engage in constructive criticism and dialogue with those in disagreement and will not engage in personal attacks or ad hominem arguments.
    The statement engages in constructive criticism without personal attacks. [+1]